在设定的半径内从表 A 计算和返回表 B 中最接近的项目的最快方法是什么

Question

I have a very large amount of lat/long coordinates in Table 1, as well as Table 2. For example, let's say there are 100,000 coordinates in both tables. I need to return the closest pair of coordinates in Table 2 from Table 1, as long as they are within a set minimum distance (let's say, 100 meters) for each unique item from Table 1 (up to 100,000 items, but then culled down to 100 meters is my expected output).

I am fairly familiar with the Geometry and Geography parts of MSSQL, and would traditionally approach the following with something like this:

Select
Table1ID = T1.ID,
Table2ID = T2.ID,
Distance = T1.PointGeog.STDistance(T2.PointGeog),
Keep = 0
into #Distance 
From #Table1 T1
   cross join #Table2 T2
where T1.PointGeog.STDistance(T2.PointGeog) <= 100

which would return all items from Table2 that are within 100 meters of Table1

Then, to limit to only the closest items, I could:

Update #Distance
 set Keep = 1
from #Distance D 
   inner join 
   (select shortestDist = min(Distance), Table1ID from #Distance GROUP BY 
    Table1ID) A
    on A.ID = D.Table1ID and A.shortestDist = D.Distance

and then delete anything where keep <> 1

This works, however it takes absolutely forever. The cross join creates an absurd amount of calculations that SQL needs to handle, which results in ~ 9 minute queries on MSSQL 2016. I can limit the range of the portion of Table 1 and Table 2 that I compare with SOME criteria, but really not much. I'm just really not sure how I could make the process quicker. Ultimately, I just need: closest item, distance from T2 to T1.

I have played around with a few different solutions, but I wanted to see if the SO community has any additional ideas on how I could better optimize something like this.

Answer 1

Try CROSS APPLY:

SELECT 
    T1.ID, TT2.ID, T1.PointGeog.STDistance(TT2.PointGeog)
FROM #Table1 as T1
CROSS APPLY (SELECT TOP 1 T2.ID, T2.PointGeog 
  FROM #Table2 as T2
  WHERE T1.PointGeog.STDistance(T2.PointGeog) <= 100
  ORDER BY T1.PointGeog.STDistance(T2.PointGeog) ASC) as TT2

Answer 2

I have played around with a new option, and I think this is the fastest I have gotten the calculation - to about 3 minutes.

I changed Table1 to be:

select
ID,
PointGeog,
Buffer = PointGeom.STBuffer(8.997741566866716e-4)
into #Table1

where the buffer is 100/111139 (convert degrees to meters)

and then

if object_id('tempdb.dbo.#Distance') is not null
drop table #Distance 
Select 
T1ID = T1.ID,
T1Geog = T1.PointGeog,
T2ID = T2.ID,
T2Geog = T2.PointGeog,
DistanceMeters = cast(null as float),
DistanceMiles = cast(null as float),
Keep = 0
Into #Distance
From #Table1 T1
    cross join #Table2 T2
Where T1.Buffer.STIntersects(T2.PointGeom) = 1

which does not calculate the distance, but first culls the dataset to anything within 100 meters. I can then pass an update statement to calculate the distance on a substantially more manageable dataset.

Answer 3

Create a spatial index on geom column on both tables and it shouldn't be too bad performance. Something like:

CREATE SPATIAL INDEX spat_t ON  [#Table1]
    (
        [PointGeog]
    )

I ran some tests with 100k dots on my laptop and it took 3 minutes to "join"